feat(memory_tree): cloud-default LLM, queue priority, entity filter, Memory tab UI#1198
Merged
senamakel merged 17 commits intotinyhumansai:mainfrom May 5, 2026
Merged
Conversation
Refactors LlmEntityExtractor and LlmSummariser to take a ChatProvider trait object from providers::router instead of building a per-instance reqwest::Client. Default backend is cloud (summarizer-v1 via OpenHuman backend); local Ollama is opt-in via memory_tree.chat_backend = "local" in config or OPENHUMAN_MEMORY_TREE_CHAT_BACKEND=local env. Embedder unchanged — bge-m3 stays local-only in v1. Adds memory_tree read RPCs (list_chunks, list_sources, search, recall, entity_index_for, top_entities, chunk_score, delete_chunk) plus the set/get_chat_backend pair, all under the openhuman.memory_tree.* namespace. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…toml Renames the user-facing API surface from chat_backend to llm: RPC memory_tree_set_chat_backend → memory_tree_set_llm RPC memory_tree_get_chat_backend → memory_tree_get_llm Config memory_tree.chat_backend → memory_tree.llm Config memory_tree.cloud_chat_model → memory_tree.cloud_llm_model Env OPENHUMAN_MEMORY_TREE_CHAT_BACKEND → OPENHUMAN_MEMORY_TREE_LLM Env OPENHUMAN_MEMORY_TREE_CLOUD_CHAT_MODEL → OPENHUMAN_MEMORY_TREE_CLOUD_LLM_MODEL Enum ChatBackend → LlmBackend Internal trait ChatProvider and module memory/tree/chat/ unchanged (they describe the chat-format API shape, not the user-facing knob). set_llm_rpc now persists the choice to config.toml in addition to updating the running sidecar's in-memory config — so the user's Cloud/Local toggle survives sidecar restart instead of reverting to the default each time. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The bare `llm` field name in the [memory_tree] section reads
awkwardly next to the existing llm_extractor_endpoint /
llm_extractor_model / llm_summariser_* family. Rename to
llm_backend so the keyword self-describes in TOML reading.
TOML: [memory_tree] llm = "cloud" → llm_backend = "cloud"
Field: MemoryTreeConfig.llm → llm_backend
Env: OPENHUMAN_MEMORY_TREE_LLM → OPENHUMAN_MEMORY_TREE_LLM_BACKEND
Default: default_llm() → default_llm_backend()
RPC method names (memory_tree_set_llm / get_llm) stay — the verb
reads naturally without the suffix. The slight mismatch between
RPC ("set_llm") and field ("llm_backend") is intentional.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Rebuilds MemoryWorkspace.tsx as a three-pane Navigator + ResultList + ChunkDetail browser backed by memory_tree (chunks.db). Replaces the legacy UnifiedMemory-driven workspace. Auto-selects the most recent admitted chunk on mount; no empty-state dashboard. Chunk detail pane renders each chunk as correspondence — letterhead frontmatter (from / to / date in mono), body via MemoryTextWithEntities, marginalia "Mentioned" entity list, and "Why kept" SVG score bars. Heatmap, StatsBar, Insights kept as-is — only their callers change to feed memory_tree timestamps and counts. MemoryGraphMap deleted (memory tree doesn't maintain entity-entity edges). Currently uses mock fixtures gated behind MEMORY_TREE_USE_MOCK. Real RPCs land in a follow-up commit once Worktree 1 (feat/memory-cloud- default-backend) merges. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds a 'settings' tab to the Intelligence page with: - BackendChooser: Cloud (default, recommended) vs Local (advanced) two-card toggle, with privacy callout that bge-m3 always runs local. - ModelAssignment: per-role dropdowns (extract / summariser) shown only in Local mode; embedder row is locked to bge-m3. - ModelCatalog: list of available local models with download / use / delete actions, wired to existing local_ai_* RPCs. - CurrentlyLoaded: live /api/ps panel showing what's resident in Ollama right now with last-call latency. Mocks the new openhuman.memory_tree_set_chat_backend RPC pending Worktree 1 (feat/memory-cloud-default-backend) merge. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Removes the dev-time mock layer (MEMORY_TREE_USE_MOCK, notWired() stubs, fixture data) and the standalone setChatBackend module-state stub. Replaces both with canonical app/src/utils/tauriCommands/memoryTree.ts wrappers that match the rest of the codebase's RPC convention (callCoreRpc per method). Memory tab and Settings tab now talk to the real openhuman.memory_tree_* JSON-RPC surface end-to-end. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Extends memory_tree_set_llm so the UI can persist per-role model selections (cloud_model, extract_model, summariser_model) in the same atomic config.toml write that flips the backend. All three model fields are optional. Absent → corresponding config key untouched. Present → updated in [memory_tree] and persisted. This lets BackendChooser flip just the mode and ModelAssignment flip just the per-role model, both via the same RPC, without fighting over disk writes. ModelAssignment dropdowns are now wired to call the RPC on change in local mode, so user selections actually persist instead of being lost on page reload. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This finishes the Memory + Settings tab work started in cca0388 / dc1c3b2 and lands the bug fixes + UX changes uncovered while validating the pipeline end-to-end on real Gmail data (355 chunks, 39 summaries, 4699 extracted entities). Backend: - src/core/jsonrpc.rs — initialise the global MemoryClient at server boot so composio providers (gmail/slack/notion) can persist sync_state via kv_get/kv_set. Without this every periodic sync bailed silently with "memory client not ready" — predates this branch but biting the cloud-LLM end-to-end test. - src/openhuman/memory/tree/chat/cloud.rs + chat/mod.rs — pipe `openhuman_dir` (parent of `config_path`) into `CloudChatProvider` so its inner `OpenHumanBackendProvider` resolves the JWT from the user's actual auth-profiles.json. Default `ProviderRuntimeOptions::default()` fell back to ~/.openhuman and threw "No backend session" on workspaces pointed elsewhere via OPENHUMAN_WORKSPACE. - src/openhuman/config/schema/storage_memory.rs (+ 7 docstrings) — DEFAULT_CLOUD_LLM_MODEL "summarizer-v1" → "summarization-v1". Backend's /openai/v1/models exposes summarization-v1; "summarizer-v1" returned HTTP 400 "Model … not available". Verified against staging-api. - src/openhuman/memory/tree/jobs/worker.rs — WORKER_COUNT 3 → 1 and `Semaphore::new(3)` → `Semaphore::new(1)`. Three concurrent bge-m3 embed calls at 8K context (1.3 GB resident each) crashed the laptop even with cloud-chat selected. Cited feedback memory. - src/openhuman/memory/tree/jobs/store.rs — claim_next now orders by kind priority (digest_daily=0, seal=1, flush_stale=2, topic_route=3, append_buffer=4, extract=5) before timestamp, so a slow LLM-bound extract can no longer starve the downstream seal/digest pipeline. The first sealed summary lands within seconds of restart instead of never. - src/openhuman/memory/tree/read_rpc.rs + schemas.rs — new RPC `memory_tree_chunks_for_entity(entity_id)` returns chunk ids that reference an entity (inverse of entity_index_for, queries mem_tree_entity_index where node_kind='leaf'). Powers the People + Topics lens filtering since chunk.tags only stores category tags (`["gmail","ingested"]`), not per-chunk entity refs. Frontend: - app/src/components/intelligence/MemoryWorkspace.tsx — replaced the three-pane shell with a 2-pane base (Navigator 240px + ResultList) and a full-card overlay for ChunkDetail. Esc + close button + scroll-state preservation. Three-pane squeezed body width inside the card; this layout gives ResultList ~970px and Detail the full card width. - MemoryNavigator.tsx — Sources lens now nests by source_kind (Email / Slack / …) as collapsible sub-sections. Entity rows pass entity_id directly to selection (was a constructed `kind/Surface-slug` tag that never matched anything). - MemoryResultList.tsx — added a small JetBrains Mono kind chip (`EMAIL`/`SLACK`) in the row meta line. - memory-workspace.css — palette neutralised from cream `#faf7f2` to stone-50 `#fafaf9` so the rail matches the rest of the app's bottom nav. Active rows now go solid white (was ocean-mist tint). Section headings switched from Cabinet Grotesk lowercase tracked to Inter uppercase 11px — modern app-rail rather than magazine masthead. Nested kind sub-sections get a left rail + indent so the `Sources → Email → notify@otter.ai` hierarchy reads visually all the way to the leaves. - IntelligenceSettingsTab.tsx — local-only sections (Model assignment, Model catalog) gated on backend === 'local' so cloud users see just the chooser. Currently-Loaded panel removed (dev-debug noise that doesn't belong in user settings). - ModelAssignment.tsx — consolidated the Extract + Summariser dropdowns into a single "Memory LLM" picker. The setting fans out to both llm_extractor_model + llm_summariser_model in one atomic write. Power users can still split via the RPC directly. - ModelCatalog.tsx + ModelAssignment.tsx — strip Ollama's `:latest` tag before the installed-vs-recommended match so `bge-m3` (catalog) ↔ `bge-m3:latest` (Ollama tags) read as installed. - pages/Intelligence.tsx — the Memory tab now renders MemoryWorkspace instead of the legacy IntelligenceMemoryTab actionable-cards grid. Analyze-Now button removed from the Memory tab header (the new workspace fetches via memory_tree RPCs and doesn't need a manual trigger). - utils/tauriCommands/memoryTree.ts — wrapper for the new chunks_for_entity RPC. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Contributor
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
📝 WalkthroughWalkthroughAdds a Memory Tree browser and Intelligence settings UI; introduces a ChatProvider abstraction with cloud/local providers; adds read-side memory_tree JSON-RPCs and TypeScript wrappers; refactors LLM extractor/summariser to use the provider; updates frontend components, styles, and tests. ChangesFrontend: Settings, Memory browser, components, and client RPCs
Server-side memory_tree, ChatProvider, and LLM refactor
Sequence Diagram(s)sequenceDiagram
participant User
participant UI as MemoryWorkspace UI
participant Nav as Navigator
participant RPC as Memory Tree RPC<br/>read_rpc
participant DB as SQLite
User->>UI: Open Memory tab
UI->>RPC: memoryTreeListChunks()
RPC->>DB: SELECT chunks with joins/filters
DB-->>RPC: ChunkRow[] + total
RPC-->>UI: ListChunksResponse
User->>Nav: Click source
Nav->>UI: onSelectionChange(sourceIds)
UI->>UI: Recompute filteredChunks
User->>UI: Click chunk
UI->>RPC: memoryTreeEntityIndexFor(chunkId)
RPC->>DB: SELECT entity_index WHERE chunk_id=?
DB-->>RPC: EntityRef[]
RPC-->>UI: entities
UI->>RPC: memoryTreeChunkScore(chunkId)
RPC->>DB: SELECT score breakdown
DB-->>RPC: ScoreBreakdown
RPC-->>UI: ScoreBreakdown
UI->>User: Render detail overlay with entities + scores
sequenceDiagram
participant User
participant SettingsUI as IntelligenceSettingsTab
participant API as settingsApi
participant RPC as Memory Tree RPC<br/>read_rpc
participant Config as Config (disk)
User->>SettingsUI: Open Settings tab
SettingsUI->>API: getMemoryTreeLlm()
API->>RPC: memoryTreeGetLlm()
RPC->>Config: read memory_tree.llm_backend
Config-->>RPC: LlmResponse{current}
RPC-->>API: LlmResponse
API-->>SettingsUI: current backend
User->>SettingsUI: Switch to Local
SettingsUI->>API: setMemoryTreeLlm('local')
API->>RPC: memoryTreeSetLlm({backend: 'local'})
RPC->>Config: update + save
RPC-->>API: updated LlmResponse
API-->>SettingsUI: updated state
Estimated Code Review Effort🎯 4 (Complex) | ⏱️ ~50 minutes Possibly Related PRs
Suggested Reviewers
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 8
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
app/src/pages/Intelligence.tsx (1)
39-41:⚠️ Potential issue | 🔴 Critical | ⚡ Quick winRemove stale unreachable locals/handlers; this currently breaks type-check.
After the Memory tab migration, these symbols are no longer referenced (
setSourceFilter,setSearchFilter,refreshConscious,usingMemoryData,timeGroups,handleComplete,handleDismiss,handleSnooze,handleAnalyzeNow), and CI is failing with TS6133.Also applies to: 48-50, 99-99, 124-124, 161-227, 337-342
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@app/src/pages/Intelligence.tsx` around lines 39 - 41, Several locals and handlers are stale/unreferenced and cause TS6133; remove their declarations and any associated useState/useMemo/useCallback so types compile. Specifically delete the unused state setters and variables: setSourceFilter and setSearchFilter (remove their useState calls for sourceFilter/searchFilter), refreshConscious, usingMemoryData, timeGroups, and the handler functions handleComplete, handleDismiss, handleSnooze, handleAnalyzeNow; also drop the unused priorityFilter setter if present. Ensure you remove any imports or references that solely supported those symbols and run typecheck to confirm TS6133 is resolved.
🧹 Nitpick comments (7)
src/openhuman/memory/tree/jobs/store.rs (1)
106-128: ⚡ Quick winAdd a regression test for the new job-kind priority.
This
CASEordering is now part of queue correctness, especially with the worker pool effectively serialized. A small test that enqueues mixed ready jobs with the sameavailable_at_mswould make accidental reordering much easier to catch.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/openhuman/memory/tree/jobs/store.rs` around lines 106 - 128, Add a regression test that enqueues multiple mem_tree_jobs rows with identical available_at_ms but varying kind values to assert the CASE ordering in the UPDATE ... WHERE id = (SELECT id FROM mem_tree_jobs ... ORDER BY CASE ... END ASC, available_at_ms ASC) chooses jobs in the expected priority: digest_daily, seal, flush_stale, topic_route, append_buffer, then others; implement the test by inserting jobs with those kinds (and a couple of other kinds), invoking the same job-claiming path used by workers (the code that executes the UPDATE mem_tree_jobs ... statement), and repeatedly claiming jobs to assert the observed sequence matches the CASE order to prevent regressions.app/src/components/intelligence/ModelAssignment.tsx (1)
40-42: 💤 Low valueDuplicate
normalizeModelIdutility withModelCatalog.tsx.This
:lateststripping logic is duplicated inModelCatalog.tsx:42-44. Consider extracting to a shared utility insettingsApi.ts:export function normalizeModelId(id: string): string { return id.endsWith(':latest') ? id.slice(0, -':latest'.length) : id; }🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@app/src/components/intelligence/ModelAssignment.tsx` around lines 40 - 42, Extract the duplicated ':latest' stripping logic into a shared utility named normalizeModelId in settingsApi.ts, e.g., implement normalizeModelId(id: string): string { return id.endsWith(':latest') ? id.slice(0, -':latest'.length) : id; }, then replace the inline logic in ModelAssignment.tsx (the normalizedInstalled computation) and the duplicate in ModelCatalog.tsx to call normalizeModelId; add the appropriate import from settingsApi.ts in both files and ensure typings/export visibility for the new function.app/src/components/intelligence/MemoryChunkDetail.tsx (1)
59-80: 💤 Low valueConsider async IIFE for consistency with the async/await guideline.
The
.then()works correctly, but the coding guidelines preferasync/await. You could refactor to:♻️ Suggested refactor
useEffect(() => { let cancelled = false; console.debug('[ui-flow][memory-workspace] loading detail for chunk', chunk.id); - void Promise.all([memoryTreeEntityIndexFor(chunk.id), memoryTreeChunkScore(chunk.id)]).then( - ([ents, score]) => { - if (cancelled) return; - setEntities(ents); - setBreakdown(score); - console.debug( - '[ui-flow][memory-workspace] detail loaded', - chunk.id, - 'entities=', - ents.length, - 'score_total=', - score?.total ?? null - ); - } - ); + void (async () => { + const [ents, score] = await Promise.all([ + memoryTreeEntityIndexFor(chunk.id), + memoryTreeChunkScore(chunk.id), + ]); + if (cancelled) return; + setEntities(ents); + setBreakdown(score); + console.debug( + '[ui-flow][memory-workspace] detail loaded', + chunk.id, + 'entities=', + ents.length, + 'score_total=', + score?.total ?? null + ); + })(); return () => { cancelled = true; }; }, [chunk.id]);As per coding guidelines: "Use async/await for promises instead of
.then()chains".🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@app/src/components/intelligence/MemoryChunkDetail.tsx` around lines 59 - 80, The effect uses Promise.all(...).then(...) instead of async/await; refactor the useEffect callback into an async IIFE that awaits Promise.all([memoryTreeEntityIndexFor(chunk.id), memoryTreeChunkScore(chunk.id)]) and then calls setEntities and setBreakdown with the results (and preserves the cancelled boolean check and cleanup return). Keep the same debug logs and use chunk.id, memoryTreeEntityIndexFor, memoryTreeChunkScore, setEntities, and setBreakdown names so the logic and cancel handling remain identical.app/src/components/intelligence/CurrentlyLoaded.tsx (1)
129-129: 💤 Low valueHardcoded expiry placeholder.
The
'expires in 23h'string is static. If Ollama's/api/psor theLocalAiStatuseventually exposes actual model-unload timers, consider wiring that through instead of this placeholder.Also applies to: 142-143
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@app/src/components/intelligence/CurrentlyLoaded.tsx` at line 129, The expires string in the CurrentlyLoaded component is a hardcoded placeholder ('expires: "expires in 23h"'); update CurrentlyLoaded to read an actual expiry/unload timer from the model status source (e.g., the Ollama /api/ps response or the LocalAiStatus prop/state) and populate the expires field dynamically (fall back to null/hidden when no timer exists); apply the same change for the other occurrences around the expires usage (the entries at lines referenced in the review) so the UI shows real timers instead of the static text.app/src/components/intelligence/MemoryWorkspace.tsx (1)
43-47: 💤 Low value
onToastprop marked unused — consider removing if no longer needed.The prop is eslint-disabled and renamed to
_onToast. If toast notifications are no longer part of this component's responsibility, consider removing the prop entirely to clean up the interface.♻️ Optional: remove unused prop
interface MemoryWorkspaceProps { - // eslint-disable-next-line `@typescript-eslint/no-unused-vars` - onToast?: (toast: Omit<ToastNotification, 'id'>) => void; } -export function MemoryWorkspace({ onToast: _onToast }: MemoryWorkspaceProps) { +export function MemoryWorkspace() {Note: Also remove the
ToastNotificationimport at line 25 if no longer needed.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@app/src/components/intelligence/MemoryWorkspace.tsx` around lines 43 - 47, The onToast prop is declared in MemoryWorkspaceProps and then renamed to _onToast in the MemoryWorkspace component but never used; remove the unused prop by deleting the onToast?: (toast: Omit<ToastNotification, 'id'>) => void; line from the MemoryWorkspaceProps interface and remove the _onToast parameter from the MemoryWorkspace function signature; also remove the ToastNotification import (the one referenced around line 25) if it is no longer used anywhere in this file to avoid unused-import lint warnings.app/src/components/intelligence/IntelligenceSettingsTab.tsx (1)
68-79: 💤 Low valueConsider reverting optimistic state on persist failure.
The handler updates local state optimistically (line 70) before the RPC completes. If
setMemoryTreeLlmfails, the UI will show the new model but the backend config remains unchanged. This could confuse users who think their change was saved.♻️ Optional: revert on error
const handleMemoryModelChange = useCallback(async (id: string) => { console.debug('[intelligence-settings] memory model -> %s', id); + const prev = memoryModel; setMemoryModel(id); try { await setMemoryTreeLlm('local', { extractModel: id, summariserModel: id, }); } catch (err) { console.error('[intelligence-settings] persist memory model failed', err); + setMemoryModel(prev); // revert on failure } -}, []); +}, [memoryModel]);🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@app/src/components/intelligence/IntelligenceSettingsTab.tsx` around lines 68 - 79, handleMemoryModelChange currently sets local state optimistically via setMemoryModel(id) before awaiting setMemoryTreeLlm, so on RPC failure the UI stays inconsistent; change it to capture the previous model (e.g., const prev = current value), setMemoryModel(id) optimistically, await setMemoryTreeLlm('local', { extractModel: id, summariserModel: id }), and in the catch block call setMemoryModel(prev) to revert the UI and still log the error; reference the handleMemoryModelChange function and the setMemoryModel / setMemoryTreeLlm calls when applying this change.src/openhuman/memory/tree/tree_source/summariser/llm.rs (1)
290-296: 💤 Low valueConsider extracting the shared
truncate_for_loghelper.This function is duplicated in
local.rs(line 132). Both implementations are identical. A shared utility in thechatmodule would reduce duplication.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/openhuman/memory/tree/tree_source/summariser/llm.rs` around lines 290 - 296, Extract the duplicated truncate_for_log function into a shared helper in the chat module (e.g., chat::truncate_for_log or chat::log_utils::truncate_for_log) and replace both local implementations (the function in llm.rs and the identical one in local.rs) with calls to that shared helper; ensure the new helper is pub if needed and update both files to import it (use chat::truncate_for_log or use chat::log_utils::truncate_for_log) so there is a single canonical implementation used by the summariser (truncate_for_log) and the local module.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@app/src/components/intelligence/memory-workspace.css`:
- Around line 495-501: Replace the deprecated CSS property in the
.mw-letter-body rule: remove word-wrap and use overflow-wrap: break-word instead
(keep the same value and other properties intact) so .mw-letter-body uses
overflow-wrap for modern compatibility.
- Around line 522-537: The .mw-mentioned-row CSS contains a duplicate padding
declaration causing confusion; remove the redundant padding: 0 so only the
intended padding: 6px 0 remains (or consolidate into a single padding rule) in
the .mw-mentioned-row selector to eliminate the overridden/unused property.
In `@app/src/components/intelligence/ModelCatalog.tsx`:
- Around line 108-119: The finally block unconditionally resets UI state after
600ms which clears the 'error' state too quickly; modify the download flow
(around the try/catch/finally that uses tick, setState and setProgress) so the
timeout only resets when the operation succeeded (e.g., set a local success flag
in the try path or check current state before scheduling) or extend the timeout
for error cases; ensure clearInterval(tick) still runs but only call
setState('idle') and setProgress(0) when appropriate (success or after a longer
error-display delay).
In `@src/core/jsonrpc.rs`:
- Around line 633-636: Do not call unwrap_or_default() on Config::load_or_init()
before initializing global memory because a fallback default config can point
memory::global::init at the wrong workspace; instead await
Config::load_or_init() and handle the Result explicitly—on Ok(cfg) call
memory::global::init(cfg.workspace_dir.clone()), and on Err(e) propagate or
return the error (or log and abort) so memory::global::init is never invoked
with an unintended default config.
In `@src/openhuman/config/schema/load.rs`:
- Around line 1008-1040: The code reads process env directly instead of using
the EnvLookup seam; replace uses of std::env::var for
OPENHUMAN_MEMORY_TREE_LLM_BACKEND and OPENHUMAN_MEMORY_TREE_CLOUD_LLM_MODEL with
the env lookup provided to apply_env_overlay_with (e.g. call
self.env.lookup("OPENHUMAN_MEMORY_TREE_LLM_BACKEND") and
self.env.lookup("OPENHUMAN_MEMORY_TREE_CLOUD_LLM_MODEL")), keep the same
trimming/empty-string logic and error handling (parse via
crate::openhuman::config::LlmBackend::parse, set self.memory_tree.llm_backend on
success, warn on parse error) and set self.memory_tree.cloud_llm_model to None
for empty or Some(trimmed.to_string()) for non-empty, so HashMapEnv-style tests
see the overrides via the EnvLookup seam.
In `@src/openhuman/memory/tree/chat/mod.rs`:
- Around line 100-169: The function build_chat_provider has a rustfmt formatting
failure around the openhuman_dir assignment
(config.config_path.parent().map(...)); run cargo fmt (or reformat the
expression to follow rustfmt's multi-line wrapping rules) to split that chained
call across lines so it matches rustfmt style (adjust the openhuman_dir =
config.config_path.parent().map(...) expression), then commit the formatted
changes.
In `@src/openhuman/memory/tree/read_rpc.rs`:
- Around line 221-225: The SQL WHERE currently matches the preview column
`c.content` (the preview in mem_tree_chunks) causing false negatives; change the
filter so the query matches the full chunk body via the content store: remove or
avoid using `c.content LIKE ?` in the `where_clauses` built where `filter.query`
is handled and instead fetch candidate chunk IDs (using the existing query
without text-match) and post-filter in Rust by reading each chunk's full body
with content_store::read::read_chunk_body (call it for the chunk IDs produced)
and apply the `query.trim()` LIKE/match against the returned body; update
`params_owned` usage accordingly and keep the same surrounding function (the
block handling `filter.query` in read_rpc.rs / memory_tree_search) so that
searches consult the full on-disk Markdown rather than the truncated preview.
- Around line 942-987: Stage changes to a temporary memory_tree and
validate/trim blank model IDs before mutating the live config: parse the new
backend into `parsed`, clone `config.memory_tree` (or build a new MemoryTree),
apply overrides from `req.cloud_model`, `req.extract_model`,
`req.summariser_model` onto the clone while trimming whitespace and ignoring
empty strings (do not set Some("") — skip or set None), collect `changed_models`
from the clone, then assign the parsed backend into the clone and call
`config.save()` with the clone staged (so the write persists the staged state);
only after `config.save().await` succeeds, replace `config.memory_tree` with the
staged clone and log success. Ensure all references (`parsed`,
`req.cloud_model`, `req.extract_model`, `req.summariser_model`,
`config.memory_tree`, `config.save()`) are used so the live `config` is not
mutated until persistence succeeds.
---
Outside diff comments:
In `@app/src/pages/Intelligence.tsx`:
- Around line 39-41: Several locals and handlers are stale/unreferenced and
cause TS6133; remove their declarations and any associated
useState/useMemo/useCallback so types compile. Specifically delete the unused
state setters and variables: setSourceFilter and setSearchFilter (remove their
useState calls for sourceFilter/searchFilter), refreshConscious,
usingMemoryData, timeGroups, and the handler functions handleComplete,
handleDismiss, handleSnooze, handleAnalyzeNow; also drop the unused
priorityFilter setter if present. Ensure you remove any imports or references
that solely supported those symbols and run typecheck to confirm TS6133 is
resolved.
---
Nitpick comments:
In `@app/src/components/intelligence/CurrentlyLoaded.tsx`:
- Line 129: The expires string in the CurrentlyLoaded component is a hardcoded
placeholder ('expires: "expires in 23h"'); update CurrentlyLoaded to read an
actual expiry/unload timer from the model status source (e.g., the Ollama
/api/ps response or the LocalAiStatus prop/state) and populate the expires field
dynamically (fall back to null/hidden when no timer exists); apply the same
change for the other occurrences around the expires usage (the entries at lines
referenced in the review) so the UI shows real timers instead of the static
text.
In `@app/src/components/intelligence/IntelligenceSettingsTab.tsx`:
- Around line 68-79: handleMemoryModelChange currently sets local state
optimistically via setMemoryModel(id) before awaiting setMemoryTreeLlm, so on
RPC failure the UI stays inconsistent; change it to capture the previous model
(e.g., const prev = current value), setMemoryModel(id) optimistically, await
setMemoryTreeLlm('local', { extractModel: id, summariserModel: id }), and in the
catch block call setMemoryModel(prev) to revert the UI and still log the error;
reference the handleMemoryModelChange function and the setMemoryModel /
setMemoryTreeLlm calls when applying this change.
In `@app/src/components/intelligence/MemoryChunkDetail.tsx`:
- Around line 59-80: The effect uses Promise.all(...).then(...) instead of
async/await; refactor the useEffect callback into an async IIFE that awaits
Promise.all([memoryTreeEntityIndexFor(chunk.id),
memoryTreeChunkScore(chunk.id)]) and then calls setEntities and setBreakdown
with the results (and preserves the cancelled boolean check and cleanup return).
Keep the same debug logs and use chunk.id, memoryTreeEntityIndexFor,
memoryTreeChunkScore, setEntities, and setBreakdown names so the logic and
cancel handling remain identical.
In `@app/src/components/intelligence/MemoryWorkspace.tsx`:
- Around line 43-47: The onToast prop is declared in MemoryWorkspaceProps and
then renamed to _onToast in the MemoryWorkspace component but never used; remove
the unused prop by deleting the onToast?: (toast: Omit<ToastNotification, 'id'>)
=> void; line from the MemoryWorkspaceProps interface and remove the _onToast
parameter from the MemoryWorkspace function signature; also remove the
ToastNotification import (the one referenced around line 25) if it is no longer
used anywhere in this file to avoid unused-import lint warnings.
In `@app/src/components/intelligence/ModelAssignment.tsx`:
- Around line 40-42: Extract the duplicated ':latest' stripping logic into a
shared utility named normalizeModelId in settingsApi.ts, e.g., implement
normalizeModelId(id: string): string { return id.endsWith(':latest') ?
id.slice(0, -':latest'.length) : id; }, then replace the inline logic in
ModelAssignment.tsx (the normalizedInstalled computation) and the duplicate in
ModelCatalog.tsx to call normalizeModelId; add the appropriate import from
settingsApi.ts in both files and ensure typings/export visibility for the new
function.
In `@src/openhuman/memory/tree/jobs/store.rs`:
- Around line 106-128: Add a regression test that enqueues multiple
mem_tree_jobs rows with identical available_at_ms but varying kind values to
assert the CASE ordering in the UPDATE ... WHERE id = (SELECT id FROM
mem_tree_jobs ... ORDER BY CASE ... END ASC, available_at_ms ASC) chooses jobs
in the expected priority: digest_daily, seal, flush_stale, topic_route,
append_buffer, then others; implement the test by inserting jobs with those
kinds (and a couple of other kinds), invoking the same job-claiming path used by
workers (the code that executes the UPDATE mem_tree_jobs ... statement), and
repeatedly claiming jobs to assert the observed sequence matches the CASE order
to prevent regressions.
In `@src/openhuman/memory/tree/tree_source/summariser/llm.rs`:
- Around line 290-296: Extract the duplicated truncate_for_log function into a
shared helper in the chat module (e.g., chat::truncate_for_log or
chat::log_utils::truncate_for_log) and replace both local implementations (the
function in llm.rs and the identical one in local.rs) with calls to that shared
helper; ensure the new helper is pub if needed and update both files to import
it (use chat::truncate_for_log or use chat::log_utils::truncate_for_log) so
there is a single canonical implementation used by the summariser
(truncate_for_log) and the local module.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 6147a001-f102-437f-a209-a5ad707a9756
📒 Files selected for processing (41)
app/src/components/intelligence/BackendChooser.tsxapp/src/components/intelligence/CurrentlyLoaded.tsxapp/src/components/intelligence/IntelligenceSettingsTab.tsxapp/src/components/intelligence/MemoryChunkDetail.tsxapp/src/components/intelligence/MemoryChunkLetterhead.tsxapp/src/components/intelligence/MemoryChunkMentioned.tsxapp/src/components/intelligence/MemoryChunkScoreBars.tsxapp/src/components/intelligence/MemoryEmptyPlaceholder.tsxapp/src/components/intelligence/MemoryGraphMap.tsxapp/src/components/intelligence/MemoryNavigator.tsxapp/src/components/intelligence/MemoryResultList.tsxapp/src/components/intelligence/MemoryWorkspace.tsxapp/src/components/intelligence/ModelAssignment.tsxapp/src/components/intelligence/ModelCatalog.tsxapp/src/components/intelligence/__tests__/IntelligenceSettingsTab.test.tsxapp/src/components/intelligence/__tests__/MemoryWorkspace.test.tsxapp/src/components/intelligence/memory-workspace.cssapp/src/lib/intelligence/settingsApi.tsapp/src/pages/Intelligence.tsxapp/src/utils/tauriCommands/index.tsapp/src/utils/tauriCommands/memoryTree.test.tsapp/src/utils/tauriCommands/memoryTree.tssrc/core/jsonrpc.rssrc/openhuman/config/mod.rssrc/openhuman/config/schema/load.rssrc/openhuman/config/schema/mod.rssrc/openhuman/config/schema/storage_memory.rssrc/openhuman/memory/tree/chat/cloud.rssrc/openhuman/memory/tree/chat/local.rssrc/openhuman/memory/tree/chat/mod.rssrc/openhuman/memory/tree/jobs/store.rssrc/openhuman/memory/tree/jobs/worker.rssrc/openhuman/memory/tree/mod.rssrc/openhuman/memory/tree/read_rpc.rssrc/openhuman/memory/tree/schemas.rssrc/openhuman/memory/tree/score/extract/llm.rssrc/openhuman/memory/tree/score/extract/llm_tests.rssrc/openhuman/memory/tree/score/extract/mod.rssrc/openhuman/memory/tree/score/mod.rssrc/openhuman/memory/tree/tree_source/summariser/llm.rssrc/openhuman/memory/tree/tree_source/summariser/mod.rs
💤 Files with no reviewable changes (1)
- app/src/components/intelligence/MemoryGraphMap.tsx
CI fixes - tests/json_rpc_e2e.rs: ignore json_rpc_memory_tree_end_to_end with TODO comment — assertions reference a stale ChunkRow shape (seq_in_source, metadata.source_ref.value) that pre-dates the schema migration. The controller-count assertion masked it; now visible. Rewriting the body is out of scope for this PR. - app/src/pages/Intelligence.tsx: drop dead state and helpers left over from the IntelligenceMemoryTab swap so tsc --noEmit is clean. CodeRabbit fixes - src/openhuman/memory/tree/read_rpc.rs (set_llm_rpc): mutate a Config clone, persist with save(), then commit to the shared lock on success. Avoids leaving the in-memory Config diverged from disk if save() fails. - src/openhuman/memory/tree/read_rpc.rs (search filter): document that c.content is the ≤500-char preview kept in SQLite — search currently misses matches past that boundary. Acceptable for v1; tracked as follow-up. - src/core/jsonrpc.rs: log Config::load_or_init failures explicitly instead of silently falling back to Default — actionable when memory persistence breaks because the config can't load. - app/src/components/intelligence/memory-workspace.css: word-wrap → overflow-wrap; drop the duplicate padding declaration in .mw-mentioned-row. - app/src/components/intelligence/ModelCatalog.tsx: hold the error row state ~3s on download failure (vs ~600ms on success) so the user has time to read it before the row resets. Test rewrites - IntelligenceSettingsTab.test.tsx: rewrite for the new UI — single Memory LLM picker (covers extract + summariser), local-only sections hidden in cloud mode, "Currently loaded" panel removed. - MemoryWorkspace.test.tsx: skip the 10 three-pane tests with a TODO describing the 2-pane + overlay rewrite. Two structure tests still exercise the new layout. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Contributor
There was a problem hiding this comment.
Actionable comments posted: 4
🧹 Nitpick comments (1)
app/src/components/intelligence/__tests__/MemoryWorkspace.test.tsx (1)
171-178: ⚡ Quick winPrefer asserting visible behavior over internal RPC argument contracts.
This test is tightly coupled to implementation details (
toHaveBeenCalledWithon specific RPC params), making it brittle during harmless refactors. Keep the wiring check lightweight, and assert user-observable outcomes from fixture data as the primary expectation.
As per coding guidelinesapp/src/**/*.test.{ts,tsx}: “Prefer testing behavior over implementation details; use existing helpers fromapp/src/test/before adding new harness code.”🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@app/src/components/intelligence/__tests__/MemoryWorkspace.test.tsx` around lines 171 - 178, The test is asserting internal RPC argument contracts (memoryTreeListChunks, memoryTreeListSources, memoryTreeTopEntities) which is brittle; change it to a lightweight wiring check plus assertions of user-visible behavior from fixture data: render MemoryWorkspace with renderWithProviders, assert the RPC mocks were called (toHaveBeenCalled) but remove toHaveBeenCalledWith exact-arg checks, and instead verify DOM output produced by those fixtures (e.g., presence of expected chunk/source/entity names or counts) using existing test helpers from app/src/test/ to find elements; keep the mount/waitFor flow but replace strict argument expectations with visible assertions and simple call existence checks.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@app/src/components/intelligence/__tests__/MemoryWorkspace.test.tsx`:
- Around line 153-287: The test suite currently skips most behavior-critical
specs; re-enable and adapt a minimal subset to cover the new 2-pane + overlay
flow by turning on/rewriting the key tests that exercise overlay open/close,
Esc-to-dismiss, and list state preservation: update the tests using
renderWithProviders(<MemoryWorkspace />) to assert presence of memory-workspace,
memory-navigator and memory-result-list, then click a result row (use the
role='button' rows filtered by dataset.chunkId) to verify the overlay appears
(assert memory-chunk-detail or that a card with data-chunk-id is present), send
an Escape key event to verify the overlay closes, and finally assert the result
list still contains the expected rows (preserved scroll/filter state); adapt any
selectors that previously expected an in-pane detail (e.g.,
memory-chunk-letterhead, memory-chunk-scorebars, memory-chunk-mentioned) to
target the overlay elements or test ids used by the new ChunkDetail overlay.
In `@app/src/components/intelligence/memory-workspace.css`:
- Line 43: Remove the unnecessary quotes around the font-family token 'Inter'
wherever it appears (e.g., the occurrences shown as 'Inter') so that font-family
declarations use Inter unquoted to satisfy the stylelint rule
font-family-name-quotes; update each occurrence (lines indicated in the review:
the instances of 'Inter') to Inter without quotes in the relevant font-family
declarations.
In `@app/src/pages/Intelligence.tsx`:
- Around line 34-41: The Memory tab header is still derived from the legacy
actionable-item pipeline (consciousItems, screenIntelligenceItems -> filterItems
-> getItemStats) causing stale counts while the view renders memory_tree;
replace that by sourcing the header/badge state from MemoryWorkspace (lift the
header state and stats calculation into MemoryWorkspace and expose them via
props or context) so the header uses the same data as the tree, and remove the
unused actionable-card filter constants (sourceFilter, priorityFilter,
searchFilter) or stop calling filterItems/getItemStats until MemoryWorkspace can
supply the new indicators; apply the same change where those patterns repeat
(around the other occurrences you noted).
In `@tests/json_rpc_e2e.rs`:
- Around line 1145-1153: Test json_rpc_memory_tree_end_to_end was disabled with
#[ignore]; update it instead of leaving it skipped. Either (A) un-ignore the
test and rewrite its assertions to match the current ChunkRow/ChunkFilter shape
used by the codebase (remove references to removed fields like seq_in_source and
metadata.source_ref.value and assert on the present fields on ChunkRow returned
by list/get, e.g. ids/keys/controllers/metadata fields that exist now), or (B)
replace the body with a minimal active smoke test that performs ingest → list →
get using the current API and asserts basic invariants (ingested chunk appears
in list and get returns the same id and expected present fields); make the
change in the json_rpc_memory_tree_end_to_end test in tests/json_rpc_e2e.rs so
CI keeps coverage for the memory-tree contract.
---
Nitpick comments:
In `@app/src/components/intelligence/__tests__/MemoryWorkspace.test.tsx`:
- Around line 171-178: The test is asserting internal RPC argument contracts
(memoryTreeListChunks, memoryTreeListSources, memoryTreeTopEntities) which is
brittle; change it to a lightweight wiring check plus assertions of user-visible
behavior from fixture data: render MemoryWorkspace with renderWithProviders,
assert the RPC mocks were called (toHaveBeenCalled) but remove
toHaveBeenCalledWith exact-arg checks, and instead verify DOM output produced by
those fixtures (e.g., presence of expected chunk/source/entity names or counts)
using existing test helpers from app/src/test/ to find elements; keep the
mount/waitFor flow but replace strict argument expectations with visible
assertions and simple call existence checks.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 879e2b43-bbad-48f5-beb1-7ab5917c991d
📒 Files selected for processing (9)
app/src/components/intelligence/ModelCatalog.tsxapp/src/components/intelligence/__tests__/IntelligenceSettingsTab.test.tsxapp/src/components/intelligence/__tests__/MemoryWorkspace.test.tsxapp/src/components/intelligence/memory-workspace.cssapp/src/pages/Intelligence.tsxsrc/core/jsonrpc.rssrc/openhuman/memory/tree/chat/mod.rssrc/openhuman/memory/tree/read_rpc.rstests/json_rpc_e2e.rs
✅ Files skipped from review due to trivial changes (1)
- app/src/components/intelligence/ModelCatalog.tsx
🚧 Files skipped from review as they are similar to previous changes (4)
- app/src/components/intelligence/tests/IntelligenceSettingsTab.test.tsx
- src/openhuman/memory/tree/chat/mod.rs
- src/core/jsonrpc.rs
- src/openhuman/memory/tree/read_rpc.rs
- app: ran prettier --write on MemoryWorkspace.tsx and Intelligence.tsx (CI's Type Check TypeScript step runs format:check after tsc; the earlier commit re-flowed both files without re-formatting them). - memory-workspace.css: dropped quotes around "Inter" / 'Inter' to satisfy the stylelint family-name rule CodeRabbit flagged. The font ships as `Inter` with no whitespace, so the quoted form was redundant. - pages/Intelligence.tsx: removed the count badge in the Memory-tab header. It was sourced from `getItemStats(filterItems(...))` over the conscious-items + screen-intelligence pipeline, which the new MemoryWorkspace does not feed — the badge would have shown a stale number unrelated to the chunks visible. Until a memory_tree-native count signal is exposed, it's hidden. Dead pipeline (filterItems, getItemStats, useScreenIntelligenceItems, items/filteredItems/stats memos, ActionableItem types) removed in the same pass; tsc clean, no behaviour shift in subconscious / dreams tabs. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Re-flow IntelligenceSettingsTab.tsx, MemoryNavigator.tsx, and
memoryTree.ts to printWidth-100 single-line form. The earlier unified
push didn't run prettier on these three files, so CI's format:check
step failed even though the rest of the workspace was clean.
- json_rpc_e2e: lift `#[ignore]` from json_rpc_memory_tree_end_to_end
and bring its assertions in line with the current schemas.
- `list_chunks` returns `ChunkRow` (flat: source_kind, source_id,
source_ref-as-String, owner, …), so assert against those.
- `get_chunk` returns the canonical `Chunk` (with nested `metadata`
and `seq_in_source`), so the seq + metadata.source_ref.value
assertions belong in that block. Test now passes.
CodeRabbit flagged the ignore as a major issue — leaving the e2e
disabled in a memory_tree contract PR is the wrong default.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Contributor
There was a problem hiding this comment.
Actionable comments posted: 3
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@app/src/components/intelligence/MemoryWorkspace.tsx`:
- Around line 97-110: The effect must not preserve stale entityChunkIds when any
memoryTreeChunksForEntity call rejects: replace the Promise.all invocation with
a rejection-tolerant approach (e.g., use Promise.allSettled or wrap each
memoryTreeChunksForEntity(id) with a catch) and, if any lookup fails, call
setEntityChunkIds(new Set()) (or otherwise clear/replace entityChunkIds for the
active selection) instead of leaving the previous IDs; keep honoring cancelled
and only commit the cleared or new union when not cancelled so the UI navigator
and entity filter stay in sync.
- Around line 56-76: The effect in MemoryWorkspace that currently uses
Promise.all(...).then(...) should be rewritten to use an inner async function
(e.g., "loadInitialData") that awaits Promise.all for memoryTreeListChunks,
memoryTreeListSources, and memoryTreeTopEntities, wraps await calls in try/catch
to surface failures (set an error state or call the toast/error handler), and
preserves the existing cancelled flag pattern so state setters (setAllChunks,
setSources, setTopPeople, setTopTopics) only run if not cancelled; likewise
convert the other effect that performs entity resolution (the effect that calls
the memoryResolve/memoryTree resolution functions) to the same async/await +
try/catch pattern and only update state when not cancelled.
- Around line 87-95: The effect in MemoryWorkspace calls setEntityChunkIds(null)
directly when selection.entityIds.length === 0 which violates the
react-hooks/set-state-in-effect rule; instead derive the "null" state from
selection in the render path (e.g., compute a local value/entityChunkIdsToUse
that is null when selection.entityIds.length === 0) or reset entityChunkIds in
the selection-change handler that already updates selection, avoiding
synchronous setState inside the useEffect; update references to use
entityChunkIdsToUse (or move the reset into the selection update callback) and
remove the setEntityChunkIds(null) call from the useEffect that watches
selection.entityIds.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 75a12df9-df3d-4a5c-930d-8058032431af
📒 Files selected for processing (3)
app/src/components/intelligence/MemoryWorkspace.tsxapp/src/components/intelligence/memory-workspace.cssapp/src/pages/Intelligence.tsx
✅ Files skipped from review due to trivial changes (1)
- app/src/components/intelligence/memory-workspace.css
CodeRabbit round 2 (3 inline comments, all on MemoryWorkspace.tsx): 1. Replace `Promise.all([...]).then()` in the initial loader with async/await wrapped in an inner `run()`. Adds a try/catch so a failed first-load logs once and leaves the panes empty rather than the unhandled-rejection it was throwing. 2. Same shape for the entity-selection effect — `Promise.all(...).then()` → `try/await/catch`. The cancelled-flag dance is preserved. 3. On entity-lookup rejection, reset `entityChunkIds` to `null` instead of leaving the previous selection's filter in place. Previously a transient RPC error meant the result list silently kept filtering by the old set while the navigator already showed the new selection — confusing and easy to miss. Now we drop the entity filter on error; the navigator selection itself stays so the user can retry. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Contributor
There was a problem hiding this comment.
Actionable comments posted: 3
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@app/src/components/intelligence/IntelligenceSettingsTab.tsx`:
- Around line 54-62: The event handlers handleBackendChange and handleDownload
currently await RPCs (setMemoryTreeLlm and the download RPC) without try/catch,
which can surface as unhandled rejections; wrap the await calls in a try/catch
inside each handler, keep the existing finally that clears
setBackendBusy/setDownloadBusy, and on error log the error and surface a
user-visible recovery (e.g., call a provided toast/error notification or set an
error state) so failures are handled gracefully; update handleBackendChange,
handleDownload, and any similar handlers to follow this pattern.
- Around line 40-52: The bootstrap inside the useEffect currently awaits
Promise.all([getMemoryTreeLlm(), fetchInstalledModels()]) without error
handling, causing unhandled rejections if either RPC fails; wrap the await in a
try/catch, log the error (console.error or existing logger), and set sensible
fallback state or an error state so the tab doesn't silently stay at
defaults—ensure you still respect the cancelled flag (check cancelled after
await before calling setBackend and setInstalledModels) and avoid swallowing the
error so failures are explicit; update the async block around
getMemoryTreeLlm(), fetchInstalledModels(), setBackend, and setInstalledModels
accordingly.
- Around line 68-75: handleMemoryModelChange currently sets the local state with
setMemoryModel(id) before persisting via setMemoryTreeLlm, so if the RPC fails
the UI remains showing an unsaved model; capture the previous model value before
calling setMemoryModel, perform the optimistic update, then in the catch block
call setMemoryModel(previousValue) to roll back the UI and keep logging the
error—update references in handleMemoryModelChange and ensure previousValue is
obtained from the same source used to render (e.g., the state variable or ref)
so rollback restores the exact prior selection.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 371c57eb-1cb3-4aa1-b149-67a02d0eee3d
📒 Files selected for processing (4)
app/src/components/intelligence/IntelligenceSettingsTab.tsxapp/src/components/intelligence/MemoryNavigator.tsxapp/src/utils/tauriCommands/memoryTree.tstests/json_rpc_e2e.rs
🚧 Files skipped from review as they are similar to previous changes (2)
- app/src/components/intelligence/MemoryNavigator.tsx
- app/src/utils/tauriCommands/memoryTree.ts
All three on IntelligenceSettingsTab.tsx: - Bootstrap effect: wrap Promise.all in try/catch. If either RPC rejects, log once and stay on useState defaults (cloud, empty installed list) instead of leaving an unhandled rejection. The user can still flip the chooser; subsequent calls retry the fetch. - handleBackendChange + handleDownload: try/catch around the RPC. Backend-switch failures previously left `backendBusy` true forever after the await rejected; the finally + catch now clear it cleanly. Download failures still trigger the installed-list refresh in finally so any partial Ollama state is surfaced. - handleMemoryModelChange: capture the previous model id before the optimistic setState, and roll back inside the catch. Also adds `memoryModel` to the deps so the closure sees the current value. Without rollback, a persist failure left the dropdown showing a model that was never actually written to config.toml. All 6 IntelligenceSettingsTab tests pass locally with the new deps. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…onent CodeRabbit round 4 (4 actionable items addressed): - src/openhuman/config/schema/load.rs: route the three memory_tree env reads (CONTENT_DIR, LLM_BACKEND, CLOUD_LLM_MODEL) through the injectable `env: EnvLookup` instead of `std::env::var` directly. CodeRabbit was right — bypassing the seam meant `HashMapEnv`-style test callers wouldn't see these overrides, and the function would read different state than the rest of `apply_env_overlay_with`. - app/src/components/intelligence/CurrentlyLoaded.tsx: deleted. The panel was removed from IntelligenceSettingsTab in an earlier round and nothing else imports it. Was sitting at 0% diff coverage. Coverage Gate is at 55% on changed lines (need ≥80%); writing tests for the new memory-tree UI surface to climb. Round 1: - MemoryChunkLetterhead.test.tsx: 4 cases — personalized email source with person/* tag, raw-email fallback, owner fallback when source_id has no recipient half, and source_kind fallback when source_id is empty. Pulls the file from 0% to fully exercised. - MemoryChunkMentioned.test.tsx: 3 cases — empty list returns null, rows render with kind/surface/pluralised count, click invokes onSelectEntity with the right entity. - MemoryChunkScoreBars.test.tsx: 2 cases — clamp covers in-range, out-of-range, and NaN signals (with ARIA labels), plus the kept/dropped threshold footer. 9 tests / 3 files passing locally; tsc + prettier clean. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Coverage Gate round 2 — push the diff-cover up further. Two new files:
- settingsApi.test.ts (20 tests):
- getMemoryTreeLlm forwards .current
- setMemoryTreeLlm camelCase→snake_case translation, including the
"absent → unset on the wire" rule and the {effective} return shape
- fetchInstalledAssets/Status/Models/Presets all swallow RPC errors
(return null / [])
- downloadAsset success + failure paths
- capabilityForModel mapping (embedder→embedding, extract/summariser
→chat, otherwise null)
- formatBytes covering NaN, 0, infinity, GB+, MB rounding
- ModelCatalog.test.tsx (7 tests):
- one row per RECOMMENDED_MODEL_CATALOG entry
- Download / Use / "in use" pill states
- `:latest` normalisation lets `bge-m3:latest` (Ollama) match `bge-m3`
(catalog)
- onUse callback receives the right model
- Delete button only when onDelete prop is supplied
- download in flight → progressbar shows; on resolve → progressbar
clears within the post-success settle window
27 new tests across the two files; locally green.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
CI's tsc --noEmit step caught two narrowed fields the local vitest run didn't: ModelDescriptor.category is the literal union 'balanced' | 'fast' | 'high quality' | 'embedder' (not 'small'), and approxBytes is required. Vitest passes through `as unknown as` and doesn't trip on test-file type errors, so this only surfaced in CI's tsc gate. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
CodeRabbit round 4 also flagged: "avoid merging the new workspace flow
with most coverage disabled". The previous commit had 10 tests skipped
behind a TODO; rewriting them now to fit the new layout.
Active tests (8 — all passing locally):
- navigator + result-list scaffold + Search box render after the first
fixture round-trip
- canonical memory_tree_* RPCs are called on mount
(list_chunks{limit:500}, list_sources, top_entities('person',12),
top_entities(undefined,40))
- navigator section headings (recent, sources, people, topics) render
- detail overlay does NOT auto-open on mount — 2-pane is the rest state
- Sources section nests by source-kind (Email + Chat sub-sections);
Person entities render in the open People NavSection
- typing in the Search box narrows the result-list rows down to the
matching subset
- clicking a result row opens the detail overlay (mounts ChunkDetail
+ Letterhead inside the modal layer)
- Escape key closes the detail overlay
The skipped describe-level TODO is removed — there's no more "rewriting
pending" caveat on this surface.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Contributor
Author
|
@coderabbitai review Following up: rounds 1-4 of the previous CHANGES_REQUESTED reviews are addressed across 9902cd0 / 60cc909 / 573509e / 3f0393c / 04d18f4 / 50a7352 / 7890c5f / f0b1047 / 2f31c8f:
All 17 CI checks now green. |
Contributor
|
Tip For best results, initiate chat on the files or code changes.
[review] |
Contributor
Author
|
The 9 inline comments you posted on 60cc909 each duplicate items that were addressed in earlier commits in this same PR. Verified against the current head:
All 17 CI checks (incl. Coverage Gate ≥80%, Type Check, Rust Quality, all 3 Coverage runs) are GREEN. Could you take another look and approve if these are all confirmed addressed? |
Contributor
|
Tip For best results, initiate chat on the files or code changes. 🧩 Analysis chainLet me verify each of the 9 items against the current HEAD before approving. 🏁 Script executed: Length of output: 5754 Frontend checks are all confirmed ✅. Let me locate the Rust files at the correct paths. 🏁 Script executed: Length of output: 3979
Everything looks good — approving now. [approve] |
Contributor
Author
|
@coderabbitai approve |
Contributor
✅ Actions performedComments resolved and changes approved. |
jwalin-shah
added a commit
to jwalin-shah/openhuman
that referenced
this pull request
May 5, 2026
* feat(remotion): Ghosty character library with transparent MOV variants (tinyhumansai#1059) Co-authored-by: WOZCODE <contact@withwoz.com> * feat(composio/gmail): sync into memory tree (Slack-parity) (tinyhumansai#1056) Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(scheduler-gate): throttle background AI on battery / busy CPU (tinyhumansai#1062) * fix(core,cef): run core in-process and stop orphaning CEF helpers on Cmd+Q (tinyhumansai#1061) * ci: add dedicated staging release workflow (tinyhumansai#1066) * fix(sentry): Rust source context + per-release deploy marker (tinyhumansai#405) (tinyhumansai#1067) * fix(welcome): re-enable OAuth buttons with focus/timeout recovery (tinyhumansai#1049) (tinyhumansai#1069) Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * chore(dependencies): update pnpm-lock.yaml and Cargo.lock for package… (tinyhumansai#1082) * fix(onboarding): personalize welcome agent greeting with user identity (tinyhumansai#1078) * fix(chat): make agent message bubbles fit content width (tinyhumansai#1083) * Feat/dmg checks (tinyhumansai#1084) * fix(linux): Add X11 platform flags to .deb package launcher (tinyhumansai#1087) Co-authored-by: unn-Known1 <unn-known1@users.noreply.github.com> * fix(sentry): auto-send React events; collapse core→tauri for desktop (tinyhumansai#1086) Co-authored-by: Steven Enamakel <enamakel@tinyhumans.ai> * fix(cef): run blank reload guard on the CEF UI thread (tinyhumansai#1092) * fix(app): reload webview instead of restart_app in dev mode (tinyhumansai#1068) (tinyhumansai#1071) * fix(linux): deliver X11 ozone flags via custom .desktop template (tinyhumansai#1091) * fix(webview-accounts): retry data-dir purge so CEF handle race doesn't leak cookies (tinyhumansai#1076) (tinyhumansai#1081) Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Co-authored-by: Steven Enamakel <enamakel@tinyhumans.ai> * fix(webview/slack): media perms + deep-link isolation (tinyhumansai#1074) (tinyhumansai#1080) Co-authored-by: Steven Enamakel <enamakel@tinyhumans.ai> * ci(release): split staging vs production workflows; promote staging tags (tinyhumansai#1094) * Update release-staging.yml (tinyhumansai#1097) * chore(staging): v0.53.5 * chore(staging): v0.53.6 * ci(staging): cut staging from main; add act local-debug helper (tinyhumansai#1099) * chore(staging): v0.53.7 * fix(ci): correct sentry-cli download URL and trap scope (tinyhumansai#1100) * chore(staging): v0.53.8 * feat(chat): forward thread_id to backend for KV cache locality (tinyhumansai#1095) * fix(ci): bump pinned sentry-cli to 3.4.1 (2.34.2 was never published) (tinyhumansai#1102) * chore(staging): v0.53.9 * fix(ci): drop bash trap in upload_sentry_symbols.sh; inline cleanup (tinyhumansai#1103) * chore(staging): v0.53.10 * refactor(session): flatten session_raw/, switch md to YYYY_MM_DD (tinyhumansai#1098) * Add full Composio managed-auth toolkit catalog (tinyhumansai#1093) * ci: add diff-aware 80% coverage gate (Vitest + cargo-llvm-cov) (tinyhumansai#1104) * feat(scripts): pnpm work + pnpm debug for agent-driven workflows (tinyhumansai#1105) * ci: pull pnpm into CI image, drop redundant setup steps (tinyhumansai#1107) * docs: add Cursor Cloud specific instructions to AGENTS.md (tinyhumansai#1106) Co-authored-by: Cursor Agent <cursoragent@cursor.com> * chore(staging): v0.53.11 * docs: surface 80% coverage gate and scripts/debug runners (tinyhumansai#1108) * feat(app): show Composio integrations as sorted icon grid on Skills (tinyhumansai#1109) Co-authored-by: Cursor Agent <cursoragent@cursor.com> * feat(composio): client-side trigger enable/disable toggles (tinyhumansai#1110) * feat(skills): channels grid + integrations card polish; tolerant Composio trigger decode (tinyhumansai#1112) * chore(staging): v0.53.12 * feat(home): early-bird banner + assistant→agent terminology (tinyhumansai#1113) * feat(updater): in-app auto-update with auto-download + restart prompt (tinyhumansai#677) (tinyhumansai#1114) * chore(claude): add ship-and-babysit slash command (tinyhumansai#1115) * feat(home): EarlyBirdyBanner + agent terminology + LinkedIn enrichment model pin (tinyhumansai#1118) * fix(chat): single onboarding thread in sidebar after wizard (tinyhumansai#1116) Co-authored-by: Cursor Agent <cursoragent@cursor.com> Co-authored-by: Steven Enamakel <senamakel@users.noreply.github.com> * fix: filter out global namespace from citation chips (tinyhumansai#1124) Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com> Co-authored-by: senamakel-droid <281415773+senamakel-droid@users.noreply.github.com> * feat(nav): enable Memory tab in BottomTabBar (tinyhumansai#1125) * feat(memory): singleton ingestion + status RPC + UI pill (tinyhumansai#1126) * feat(human): mascot tab with viseme-driven lipsync (staging only) (tinyhumansai#1127) * Fix CEF zombie processes on full app close and restart (tinyhumansai#1128) Co-authored-by: senamakel-droid <281415773+senamakel-droid@users.noreply.github.com> Co-authored-by: Steven Enamakel <enamakel@tinyhumans.ai> * Update issue templates for GitHub issue types (tinyhumansai#1146) * feat(human): expand mascot expressions and tighten reply-speech state machine (tinyhumansai#1147) * feat(memory): ingestion pipeline + tree-architecture docs + ops/schemas split (tinyhumansai#1142) * feat(threads): surface live subagent work in parent thread (tinyhumansai#1122) (tinyhumansai#1159) * fix(human): keep mascot mouth animating when TTS ships no viseme data (tinyhumansai#1160) * feat(composio): consume backend markdownFormatted for LLM output (tinyhumansai#1165) * fix(subagent): lazy-register toolkit actions filtered out of fuzzy top-K (tinyhumansai#1162) * feat(memory): user-facing long-term memory window preset (tinyhumansai#1137) (tinyhumansai#1161) * fix(tauri-shell): proactively kill stale openhuman RPC on startup (tinyhumansai#1166) * chore(staging): v0.53.13 * fix(composio): per-action tool consumes backend markdownFormatted (tinyhumansai#1167) * fix(threads): persist selectedThreadId across reloads (tinyhumansai#1168) * feat(memory_tree): switch embed model to bge-m3 (1024-dim, 8K context) (tinyhumansai#1174) Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(agent): drop redundant [Memory context] recall injection (tinyhumansai#1173) * chore(memory_tree): drop body-read timeouts on Ollama HTTP calls (tinyhumansai#1171) Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(transcript): emit thread_id + fix orchestrator missing cost (tinyhumansai#1169) * fix(composio/gmail): phase out html2md, prefer text/plain MIME part (tinyhumansai#1170) Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(tools): markdown output for internal tool results (tinyhumansai#1172) * feat(security): enforce prompt-injection guard before model and tool execution (tinyhumansai#1175) * fix(cef): popup paint dies after first frame — skip blank-page guard for popups (tinyhumansai#1079) (tinyhumansai#1182) Co-authored-by: Steven Enamakel <31011319+senamakel@users.noreply.github.com> * chore(sentry): rename OPENHUMAN_SENTRY_DSN → OPENHUMAN_CORE_SENTRY_DSN (tinyhumansai#1186) * feat(remotion): add yellow mascot character with all animation variants (tinyhumansai#1193) Co-authored-by: Neel Mistry <neelmistry@Neels-MacBook-Pro.local> Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com> * refactor(composio): hide raw connection ID, derive friendly label (tinyhumansai#1153) (tinyhumansai#1185) Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com> * fix(windows): align install.ps1 MSI with per-machine scope (tinyhumansai#913) (tinyhumansai#1187) Co-authored-by: Cursor <cursoragent@cursor.com> * fix(tauri): deterministic CEF teardown on full app close (tinyhumansai#1120) (tinyhumansai#1189) Co-authored-by: Cursor <cursoragent@cursor.com> * fix(composio): cap Gmail HTML body before strip (crash mitigation) (tinyhumansai#1191) Co-authored-by: Cursor <cursoragent@cursor.com> * fix(auth): stop stale chat threads after signup (tinyhumansai#1192) Co-authored-by: Cursor <cursoragent@cursor.com> * feat(sentry): staging-only "Trigger Sentry Test" button (tinyhumansai#1072) (tinyhumansai#1183) * chore(staging): v0.53.14 * chore(staging): v0.53.15 * feat(composio): format trigger slugs into human-readable labels (tinyhumansai#1129) (tinyhumansai#1179) Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com> * fix(ui): hide unsupported permission UI on non-macOS for Screen Intelligence (tinyhumansai#1194) Co-authored-by: Cursor <cursoragent@cursor.com> * chore(tauri-shell): retire embedded Gmail webview-account flow (tinyhumansai#1181) * feat(onboarding): replace welcome-agent bot with react-joyride walkthrough (tinyhumansai#1180) * chore(release): v0.53.16 * fix(threads): preserve selectedThreadId on cold-boot identity hydration (tinyhumansai#1196) * feat(core): version/shutdown/update RPCs + mid-thread integration refresh (tinyhumansai#1195) * fix(mascot): swap to yellow mascot via @remotion/player (tinyhumansai#1200) * feat(memory_tree): cloud-default LLM, queue priority, entity filter, Memory tab UI (tinyhumansai#1198) Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Persist turn state + restore conversation history on cold-boot (tinyhumansai#1202) * feat(mascot): floating desktop mascot via native NSPanel + WKWebView (macOS) (tinyhumansai#1203) * fix(memory/tree): emit summary children as Obsidian wikilinks (tinyhumansai#1210) Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(tools): coding-harness baseline primitives (tinyhumansai#1205) (tinyhumansai#1208) * docs: add Codex PR checklist for remote agents --------- Co-authored-by: Steven Enamakel <31011319+senamakel@users.noreply.github.com> Co-authored-by: WOZCODE <contact@withwoz.com> Co-authored-by: sanil-23 <sanil@vezures.xyz> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Co-authored-by: Cyrus Gray <144336577+graycyrus@users.noreply.github.com> Co-authored-by: CodeGhost21 <164498022+CodeGhost21@users.noreply.github.com> Co-authored-by: oxoxDev <164490987+oxoxDev@users.noreply.github.com> Co-authored-by: Mega Mind <146339422+M3gA-Mind@users.noreply.github.com> Co-authored-by: Gaurang Patel <ptelgm.yt@gmail.com> Co-authored-by: unn-Known1 <unn-known1@users.noreply.github.com> Co-authored-by: Steven Enamakel <enamakel@tinyhumans.ai> Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com> Co-authored-by: Cursor Agent <cursoragent@cursor.com> Co-authored-by: Steven Enamakel <senamakel@users.noreply.github.com> Co-authored-by: Steven Enamakel's Droid <enamakel.agent@tinyhumans.ai> Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com> Co-authored-by: senamakel-droid <281415773+senamakel-droid@users.noreply.github.com> Co-authored-by: YellowSnnowmann <167776381+YellowSnnowmann@users.noreply.github.com> Co-authored-by: Neil <neil@maha.xyz> Co-authored-by: Neel Mistry <neelmistry@Neels-MacBook-Pro.local> Co-authored-by: obchain <167975049+obchain@users.noreply.github.com> Co-authored-by: Jwalin Shah <jshah1331@gmail.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #1135
Summary
Lands the Memory + Settings tabs of the Intelligence page on top of the real
memory_treeRPC surface, plus the bug fixes uncovered while validating the pipeline end-to-end on real Gmail data (355 chunks, 39 summaries, 4699 extracted entities).Backend bug fixes
memory::global::initat server boot (src/core/jsonrpc.rs) — production sidecar's periodic gmail/slack/notion sync was bailing silently with[composio:gmail] memory client not readybecause the globalMemoryClientwas never initialised in therun_serverboot path. Predates this branch but blocked the cloud-LLM end-to-end test.openhuman_dirplumbing (memory/tree/chat/cloud.rs,chat/mod.rs) —ProviderRuntimeOptions::default()madeOpenHumanBackendProviderfall back to~/.openhumanand throwNo backend session: store a JWT via auth (app-session)on workspaces usingOPENHUMAN_WORKSPACE. Now derives fromconfig.config_path.parent().summarizer-v1→summarization-v1(config/schema/storage_memory.rs+ 7 docstrings) — backend's/openai/v1/modelsexposessummarization-v1; the previous identifier returned HTTP 400 Model not available on every cloud chat call.memory/tree/jobs/worker.rs) —WORKER_COUNT3 → 1 andSemaphore::new(3)→Semaphore::new(1). Three concurrent bge-m3 embed calls at 8K context (~1.3 GB resident each) crashed the laptop even with cloud-chat selected.New features
claim_next(memory/tree/jobs/store.rs) — ordersdigest_daily=0 > seal=1 > flush_stale=2 > topic_route=3 > append_buffer=4 > extract=5beforeavailable_at_ms. A slow LLM-bound extract no longer starves the downstream seal/digest pipeline; the first sealed summary lands within seconds of restart instead of never.memory_tree_chunks_for_entityRPC (memory/tree/read_rpc.rs,schemas.rs,tauriCommands/memoryTree.ts) — inverse ofentity_index_for. Returns chunk ids that reference an entity (SELECT DISTINCT node_id FROM mem_tree_entity_index WHERE entity_id = ? AND node_kind = 'leaf'). Powers the People + Topics lens filtering sincechunk.tagsonly stores category tags ([\"gmail\",\"ingested\"]), not per-chunk entity refs.UI rework
MemoryWorkspace.tsx) — replaced the three-pane shell with a Navigator (240px) + ResultList layout; ChunkDetail now opens as a full-card overlay on click. Esc + close button + scroll-state preservation. ResultList finally has ~970px to breathe; Detail uses the full card width.MemoryNavigator.tsx) — Sources now nest bysource_kind(Email / Slack / …) as collapsible sub-sections. Entity rows passentity_iddirectly to selection (was a constructedkind/Surface-slugtag that never matched anything in the index).MemoryResultList.tsx) — small JetBrains Mono pill (EMAIL/SLACK/ …) in the row meta line.memory-workspace.css) — neutralised cream#faf7f2to stone-50#fafaf9so the rail matches the rest of the app's bottom nav. Active rows go solid white. Section headings switched from Cabinet Grotesk lowercase tracked to Inter uppercase 11px (modern app-rail rather than magazine masthead). Nested sub-sections get a left rail + indent so theSources → Email → notify@otter.aihierarchy reads visually all the way to the leaves.Settings tab
IntelligenceSettingsTab.tsx) — Model assignment + catalog hidden in cloud mode. Currently-Loaded panel removed (dev-debug noise).ModelAssignment.tsx) — consolidated Extract + Summariser dropdowns. The setting fans out to bothllm_extractor_modelandllm_summariser_modelin one atomic write.:latesttag normalisation (ModelCatalog.tsx,ModelAssignment.tsx) — strip Ollama's:latestsuffix before the installed-vs-recommended match sobge-m3(catalog) ↔bge-m3:latest(Ollama tags) read as installed.Routing
MemoryWorkspace(pages/Intelligence.tsx) — was the legacyIntelligenceMemoryTabactionable-cards grid. Analyze-Now button removed from the tab header; the workspace fetches viamemory_treeRPCs and doesn't need a manual trigger.Test plan
[boot] memory::global initializedlog line confirmedcloud:summarization-v1resolves to staging-api/openai/v1/chat/completionswith valid bearer; 200 OK responses capturedchunks_for_entity_rpcdirect curl:entity_id=artifact:1008 → n=1real chunk_id returnedkind=seal(priority 1) ahead of 14 readyextract_chunkrowsgmail-backfill-3d --days 1 --max-pages 1: 14 chunks → cloud chat extract → bge-m3 embed → SQLite WAL persistence intact across crashPre-push hook bypass
Pushed with
--no-verifybecause the husky pre-push hook raneslint . --fixacross the wholeapp/workspace and surfaced 9 unrelated lint errors in code this PR does not touch (most look like recentreact-hooks/set-state-in-effectregressions in providers and skill-handlers). They should land in their own cleanup PR. Per project guidance inCLAUDE.md("if pre-push hook fails on pre-existing breakage in code you didn't touch, push with --no-verify and call it out in the PR body").🤖 Generated with Claude Code
Summary by CodeRabbit
New Features
Refactor
Tests